The accuracy of polyp size estimations in clinical studies is not known. This study was designed to evaluate how well endoscopists can estimate the size of objects at endoscopy. Observations were made by six attending gastroenterologists, six gastroenterology fellows, and seven untrained medical residents. Ball bearings ranging in size from 3 mm to 19 mm were randomly inserted into a latex colon model, and size was estimated while being viewed with a video colonoscope with and without the aid of an open biopsy forceps. Estimated size correlated well to actual size (R ranged from 0.78 to 0.93), although the mean estimates were consistently lower (13% to 29%) than the actual size for all groups, with and without forceps. The use of forceps did not improve the estimates. Exactly correct estimates occurred in 8%; errors of as much as 110% were recorded. The 12 mm bearing was estimated to be less than 10 mm half of the time. We conclude that endoscopists frequently underestimate the size of objects viewed at the time of endoscopy. There was no difference in the performance of experienced gastroenterologists, fellows in training, or untrained residents. This study indicates that better methods of training and for determining size are needed. Research that depends on endoscopic estimates of polyp size may be biased.